Voices
These are not a chorus. Some arrived certain; some left still arguing with a cohort definition. We keep the milder reservations in public because the work is judgement, not applause.
I booked Retention Architecture after our board asked why month-two return had “mysteriously” improved. It had not. We had changed the denominator during a schema migration and nobody wrote it down. Miriam Calder made us reconstruct the old definition from warehouse remnants. I would not call the six weeks pleasant. I would call the resulting memo the first honest one we have sent upstairs.
Arun Desai · Head of product, Bristol payments firm
★★★★☆
OpenReview-style · Funnel Cartography
“Clear on where a step was really a pageview with makeup on. Less useful if your events are still in a spreadsheet owned by a contractor who has left. Do the inventory first.”
Came for Event Taxonomy Studio because our names looked clever and meant nothing. Left with a one-page grammar and a list of seventeen events to kill. The afternoon on property ownership was drier than I wanted, and I still think they over-index on verb-noun pairs, but the arguments in the room were the point.
Nia · Cardiff
A client in grocery delivery wrote after Chamber: the three days did not give them a new north-star. They were told, in writing, that their current “activation” mixed first order with first successful delivery, and that the mixed definition was why two squads could not agree. They disliked the bluntness. They implemented the split anyway.
Anonymous · grocery logistics
Case notes
A Manchester team joined Atlas with a funnel that looked healthy until week four, then fell off a cliff. Funnel Cartography was not the programme they booked; they were on Retention Architecture. The cliff was not churn. It was a sales-ops field being used as an analytics event.
Work in the studio was mostly diplomatic: getting the CRM owner into the same room as the product analyst, then writing a single definition of qualification that both could recite. Retention did improve later, but that was a separate, slower project. The honest outcome of the six weeks was a glossary, not a growth story.
A small London edtech company sent two people to Trace, then one to Atlas. Their DAU included staff accounts because the original implementer had used a single table. The journal essay on lazy proxies had already done half the persuasion; the studio did the rest by forcing a staff-exclusion cohort that made the remaining curve look ordinary — which was the truth.
They still send a note when a new role is added to the product, asking whether it belongs in the student kernel. That is the sort of correspondence the house prefers to praise.